Muse Image: Meta’s New Image Generator Processes Public Instagram Images
Meta has unveiled Muse Image, its first in-house AI image generation model. It competes with OpenAI and Google—but is sparking discussion primarily due to a feature that, by default, makes public Instagram profiles available for other users’ AI images.
On Tuesday, Meta released Muse Image, its first proprietary image generation model. It was developed by Meta Superintelligence Labs (MSL) under the leadership of AI chief Alexandr Wang. For the department, this is the second major release following the Muse Spark language model introduced in April, which replaced the existing Llama family. According to reports, the project ran internally under the codename “Mango.”
This move is also strategic for Meta: image and video functions in the Meta AI app have previously relied on third-party providers such as Midjourney and Black Forest Labs. With Muse Image, the company aims to bring this work in-house and reduce its dependence on external models. The announcement was well-received on the stock market—Meta shares rose by approximately three percent on the day of the unveiling.
What Muse Image can do
Muse Image generates images from text descriptions, edits existing photos, removes objects, and, according to Meta, can merge multiple image templates into a single composition. The company also highlights that the model renders text legibly in graphics—an area where previous image generators frequently failed—and can generate infographics or functional QR codes. Users can also draw changes directly onto the image instead of formulating new prompts.
Technically, Meta describes the model as “agentic”: Muse Image works in conjunction with Muse Spark, plans the image composition before the actual generation, accesses web search and coding tools as needed, and performs multi-step revisions of its own results. In internal benchmarks, Meta places the model’s overall performance behind OpenAI’s GPT Image 2, but ahead of Google’s Nano Banana 2 in editing tasks involving one or more photos. These figures are from Meta’s own tests and have not yet been independently verified.
Muse Image is initially available in the Meta AI app and at meta.ai, in Instagram Stories for now in the USA, and in WhatsApp chats in selected countries. Furthermore, there are more than 30 new AI effects for Instagram. Facebook and Messenger are expected to follow later. According to Meta, use is free for “everyday creating”; those who want higher limits require one of the subscriptions introduced in May. In the coming weeks, advertisers are also expected to be able to access the model via the Advantage+ system.
Public Instagram profiles enabled by default
A different feature is causing the most stir: users can mention a public Instagram account via @-mention in a prompt, after which Meta AI uses publicly visible photos from that profile to generate a new image featuring the respective person. Meta promotes this as a way to create, for example, personalized invitations or collaborative creative concepts.
Criticism is primarily directed at the fact that the function is enabled by default for public accounts—those affected must actively turn it off (opt-out). Meta’s help section essentially states that others can create content using your own Instagram content via AI functions as long as the account is public and the default settings are active. Additionally, users are not notified when their content is used in this way. Media outlets such as WIRED, TechCrunch, and Digital Trends have pointed out that an older, archived version of this help page did not yet contain the AI-related phrasing.
Those who want to prevent this without making their account private must deactivate the option in the Instagram app under Profile → Menu → “Sharing and Reposting” that allows the use of one’s own content for AI functions at Meta—separately for posts and reels. Crucially: turning this off only applies to future generations. Previously created AI images remain and will not be deleted even if the profile is subsequently set to private.
In response, Meta points to its own protective measures. According to the company, all images generated with Muse Image carry an invisible watermark called “Content Seal” intended to identify the AI origin; furthermore, filters against clearly harmful content, such as depictions of child abuse, are built-in. Critics counter that while a retrospective watermark proves the origin of an image, it does not prevent such an image from being created in the first place.
Regulatory questions in Europe
For the European market, data protection is likely to be particularly relevant. Since the function uses publicly accessible photos of real people as a basis, observers expect close scrutiny regarding the General Data Protection Regulation (GDPR) and potential questions about the protection of biometric data. Meta has not announced any GDPR-specific adjustments for the launch. For context, some media outlets note that Meta discontinued its facial recognition system in 2021 following legal and regulatory pressure.
Muse Image is also just the beginning: With Muse Video, Meta has announced a video generation model that, according to the company, is currently in development—the company has not yet named a specific launch date.
